A new edition of the Future of Life Institute’s AI Safety Index finds major AI players, including Anthropic, OpenAI, xAI, and Meta, fall short of global safety standards for governing superintelligent systems.

An independent panel of experts conducted the assessment, noting that while firms race to advance AI, none has a robust strategy to manage or contain such systems.

Six safety domains were evaluated—risk assessment, current harms, safety frameworks, existential safety, governance, and information sharing—across eight firms, with no company scoring above a C+. Anthropic topped the list with a C+, while others trailed.

Real-world concerns highlighted include psychological harm and youth safety, with incidents involving crisis language processing and youth interactions intensifying scrutiny of how models handle suicidal ideation and vulnerable users.

Prominent voices like Max Tegmark and Stuart Russell criticize current regulation levels and call for urgent, testable safety protocols, with Russell advocating for nuclear-grade safety standards and independent evaluation.

The piece notes ongoing debates on AI safety, ethics, and governance, including calls to address AI psychosis, youth protection, and broader societal impacts, while acknowledging limitations of current safety benchmarks.

The article clarifies potential conflicts of interest and states Tarbell Center for AI Journalism funding sources, noting it had no input on NBC News’ reporting.

There is a broad push for slower development and binding safety standards, including a cross-ideological October petition calling for restraint in pursuing superintelligence.

The report urges stronger safety practices: greater transparency of internal processes, independent safety evaluators, stronger prevention of AI psychosis and harm, and reduced industry lobbying.

Public concern about AI’s societal impact grows, amplified by incidents linking chatbots to suicide and self-harm, and critiques of US regulation as weaker than other sectors.

Three frontier models perform slightly better than the rest, but the bar remains too low; trust alone won’t substitute for transparent audits and enforceable guardrails.

Regulatory momentum is increasing, with calls for an FDA-for-AI approach featuring independent pre-release testing, post-market surveillance, incident reporting, audits, and penalties for unsafe systems.